What AI Detector Do Colleges Actually Use? (2026)

Between October 2025 and April 2026, Turnitin ran its AI writing indicator across 33 million submissions from US higher education alone. Just under one in five of those papers returned with 80% or more of the text flagged as AI-written.

So the short answer to the question is Turnitin, by a wide margin and mostly for an unglamorous reason: it is already sitting inside the box that students upload to.

The longer answer is that the list of universities which have switched the feature off now includes Vanderbilt, the University of Pittsburgh, Johns Hopkins, the University of Illinois Urbana-Champaign and, since the first of January, Curtin. The question of which AI detector your college uses is two questions: what does the institution license and what has it left turned on.

Turnitin is the default because it is already in the gradebook

No university procurement team went shopping for an AI detector in 2023. However, Turnitin, which most institutions had licensed years earlier for text matching, added an AI writing indicator to a product already wired into Canvas, Blackboard, Moodle and Brightspace. The score appears in the same similarity report that the instructor was opening anyway.

That is the whole mechanism of its dominance. There was no purchase, no training and no decision required for adoption. Departments had to find another opinion if they wanted one but the default was provided for them automatically.

The same logic explains why the switch-offs are administrative rather than technical. The indicator can be turned off for an institution but that is the only control that an institution has, as Turnitin does not allow instructors to opt out of a feature that the institution has enabled.

What Turnitin's own numbers show

Turnitin publishes periodic data on what its detector reports. The Q2 2026 Learning Integrity Insights report, released on 17 July 2026 and covering the period of October 2025 to April 2026, provides the clearest picture of the scale:

• 33 million submissions from US higher education, 3.1 million from Australia, 2.1 million from the UK
• 19.4% of US higher-education submissions flagged at 80% or more AI writing, against 10.2% in Australia and 9.8% in the UK
• Roughly half of all higher-education submissions in each of the three countries showed some AI writing
• In K-12, only 5% to 6% of submissions hit the 80% mark, about half the higher-education rate

Two things are worth separating here. The volume figures are a reasonable measure of how much of the sector still has the feature enabled, since a disabled indicator produces no score. The percentages, however, are not a measure of how many students used AI. Instead, they are a measure of what the detector said, which is a different claim and Turnitin's own guidance on the limits of that claim is more careful than the way the numbers usually get quoted.

The universities that turned it off

The first to write an explanation was Vanderbilt University. On 16 August 2023, the Brightspace team at the university disabled the Turnitin detector and performed the necessary calculations out loud. Turnitin had claimed a 1% false positive rate; Vanderbilt submitted 75,000 papers in 2022 alone; 1% of 75,000 is 750 papers wrongly flagged in a single year. Vanderbilt also cited findings that detectors flag papers written by non-native English speakers as AI-written more often; Turnitin has refused to explain how the detector works beyond stating that it looks for patterns typical of AI-written text.

The University of Pittsburgh's Teaching Center is blunter still. The Center states that current AI detection software "is not yet reliable enough to be deployed without a substantial risk of false positives", that the AI detection tool has been disabled within Turnitin and that the University "does not endorse or support the use of any AI-detection tools". Not Turnitin's. Any.

Johns Hopkins disabled the feature over the same false-positive concerns. According to the teaching site for the university, there is no official policy regarding the use of AI detection software within the university. The University of Illinois Urbana-Champaign's Program in Professional Writing, however, takes it a step further in their policy: "Please do NOT use Turnitin to detect AI-generated writing."

Curtin University in Perth has made the clearest statement about where the boundary sits. From 1 January 2026, the AI writing detection feature in Turnitin will be disabled at Curtin University. In the university's words, "the regular text-matching originality checks will remain active". The stated reason is "fostering trust and clarity within a modern academic culture".

That distinction is important for anyone who reads a similarity report. Text matching and AI detection are two separate products within the same interface and a university can use one without the other.

The rest of the shortlist

Where the institution has switched Turnitin's indicator off or never had it, individual instructors and departments buy their own. These are the tools that actually turn up, with the numbers each vendor publishes about itself.

Treat every figure in that middle column as marketing until someone independent reproduces it. The accuracy of a detector on a corpus of clean GPT-3.5 essays tells you very little about its accuracy on a real third-year paper that was drafted by a student, tidied up by a model and then edited again.

Originality.ai says as much on its own product page: "An AI score shouldn't be used alone for academic integrity discipline."

What the peer-reviewed testing actually found

Three studies are worth knowing about, as they are often cited in opposite directions.

William Walters tested 16 detectors on 126 documents for Open Information Science in 2023: 42 essays from ChatGPT-3.5, 42 from ChatGPT-4 and 42 from students. Turnitin and Copyleaks correctly classified all 126 documents. This is the study that Turnitin cites on its marketing pages. It is a real result on 2023 models and unedited documents.

Weber-Wulff and colleagues ran a larger comparison of detection tools in the same year that covered 12 public tools and two commercial systems. However, their findings were the opposite of those of the previous study: the available detection tools are "neither accurate nor reliable" in their detection of AI-generated text and they tend to incorrectly classify AI-generated text as being written by humans, with performance that degrades sharply once the text has been obfuscated or paraphrased.

The most recent of the three studies is from February 2026. Hadra, Cambridge and Mesbah at Sultan Qaboos University tested Turnitin and Originality against 192 texts. These texts included writing from authentic EFL students, professionals, AI and hybrid texts. Turnitin had an accuracy of 0.61 while Originality had an accuracy of 0.69. However, on the hybrid texts, which is the type of text that most students write today, recall for Originality was near zero. The authors conclude that neither of these tools is reliable enough to be used in high-stakes decisions and that they should be used as supplements to human judgement.

Nothing in those three results is contradictory when you think about what each of them measured. Detectors are good at synthetic text from an older model but poor at mixed text from a current model.

The asterisk under 20%

The size of the percentage of AI writing in a document impacts what the score means. Turnitin's own product guidance states that "in cases where we detect less than 20% of AI writing in a document, there is a higher incidence of false positives," which is why the product prints an asterisk next to any percentage below that line. The company puts its document-level false-positive rate under 1% only for documents above 20% and its sentence-level rate at around 4%, meaning roughly four in every hundred individually highlighted sentences are human writing.

A 9% AI score with an asterisk is not weak evidence of anything. This is the part of the range that the vendor itself has flagged as unreliable. Our longer breakdown of how Turnitin's AI detection works goes through the scoring in more detail, and why an AI detector flags an essay that isn't AI covers the false-positive side.

What to assume if you are the one submitting

Assume that the indicator is on. Most institutions have not turned off the indicator, and the ones that have rarely announce it to students.

Assume that your instructor cannot see a methodology either. The score arrives without an explanation. This is exactly why a good academic integrity process will treat the score as a prompt to ask you questions rather than as a finding.

Keep the evidence of your process. The version history in Google Docs or Word, the dated drafts, the notes and the outlines are worth more in a meeting than any counter-test and some detectors, including GPTZero, now sell a playback feature that reconstructs how a document was typed.

Check if your LMS even runs one. Canvas does not have its own AI detector; what you see there comes from whatever your institution has plugged in.

If you are on the other side of this debate, accused on the strength of a percentage, then our page on AI detector false positives covers what the research supports and what it does not.

The short version

Turnitin is what most US colleges use because it came free with a product they already had. Somewhere between a handful and a long list of universities have disabled it, including Vanderbilt, Pittsburgh, Johns Hopkins, Illinois and Curtin. The reasons for this are public and specific. The alternatives that fill the gap at the department level are GPTZero, Copyleaks, Originality.ai and Pangram, all of which claim accuracy above 97% but have not been independently verified at that level. The peer-reviewed studies on the topic show that detectors are very good at detecting machine-written texts but very bad at detecting texts that are a mix of human and AI writing, which is the case with most student work today. So, a score is a prompt for a conversation, not a verdict.

If you want the longer treatment:

• How Turnitin's AI detection actually works — the scoring, the thresholds and the limits
• Why an AI detector says your essay is AI when it isn't — the false-positive mechanics
• Can AI detectors detect Claude, ChatGPT and Gemini? — model by model
• Does Canvas have an AI detector? — what your LMS is and is not doing
• How AI detectors work — perplexity, classifiers and watermarks, and which one judges your essay
• Our own AI detector — free, and it will tell you what a detector sees before your instructor does